Merge pull request #529 from NERSC/operator-env-passthru

pass extra env vars through to operators

Author: swelborn

backend/agent/.env.example

@@ -11,6 +11,7 @@ ZMQ_BIND_INTERFACE="lo"
 AGENT_NETWORKS='["perlmutter"]'
 AGENT_NAME="SecretAgentMan"
 NATS_STREAM_STORAGE_TYPE=file
+OPERATOR_EXTRA_ENV='{"CUDA_VISIBLE_DEVICES":"0","MY_FLAG":"enabled"}'
 
 # Vector log aggregation
 VECTOR_AGGREGATOR_ADDR=host.containers.internal:6000

backend/agent/interactem/agent/agent.py

@@ -93,7 +93,10 @@ def log_hook(details: stamina.instrumentation.RetryDetails) -> None:
 
 set_on_retry_hooks([log_hook])
 
-GLOBAL_ENV = {k: str(v) for k, v in cfg.model_dump().items()}
+GLOBAL_ENV = {
+    k: str(v) for k, v in cfg.model_dump(exclude={"OPERATOR_EXTRA_ENV"}).items()
+}
+GLOBAL_ENV.update(cfg.OPERATOR_EXTRA_ENV)
 GLOBAL_ENV["NATS_SERVER_URL"] = GLOBAL_ENV["NATS_SERVER_URL_IN_CONTAINER"]
 
 OPERATOR_CREDS_TARGET = "/operator.creds"

backend/agent/interactem/agent/config.py

@@ -46,6 +46,7 @@ class Settings(BaseSettings):
     VECTOR_AGGREGATOR_ADDR: str | None = None
     LOG_DIR: Path = Path("~/.interactem/logs").expanduser().resolve()
     VECTOR_CONFIG_PATH: Path | None = None
+    OPERATOR_EXTRA_ENV: dict[str, str] = Field(default_factory=dict)
 
     @model_validator(mode="after")
     def ensure_operator_creds_file(self) -> "Settings":